The invention discloses a method of identifying damage to a large structure based on three-level data fusion, comprising the following steps: A, collecting displacement and acceleration information of a large structure to be detected, and carrying out data processing; B, fusing the information collected in step A through a three-level data fusion algorithm; C, identifying the result obtained in step B, and getting the detection result through numerical simulation. Transmission of useless data and transmission of redundant information are reduced. A three-level fusion strategy is designed to complete identification of damage to a structure. First, information collected by single sensors is fused in the data level through a consistent fusion algorithm in order to improve the precision of data acquisition and reduce the amount of data transmitted by nodes. Then, damage is identified preliminarily by using an ACGA-BP neural network and taking the displacement of static measurement data and the inherent frequency of dynamic measurement data as the input parameters of the network. Finally, the identification result based on static measurement data and the identification result based on dynamic measurement data are fused based on the D-S evidence theory, and thus, the final identification result is more accurate.